Ultrafast neuromorphic photonic image processing with a VCSEL neuron.

The ever-increasing demand for artificial intelligence (AI) systems is underlining a significant requirement for new, AI-optimised hardware. Neuromorphic (brain-like) processors are one highly-promising solution, with photonic-enabled realizations receiving increasing attention. Among these, approaches based upon vertical cavity surface emitting lasers (VCSELs) are attracting interest given their favourable attributes and mature technology. Here, we demonstrate a hardware-friendly neuromorphic photonic spike processor, using a single VCSEL, for all-optical image edge-feature detection. This exploits the ability of a VCSEL-based photonic neuron to integrate temporally-encoded pixel data at high speed; and fire fast (100 ps-long) optical spikes upon detecting desired image features. Furthermore, the photonic system is combined with a software-implemented spiking neural network yielding a full platform for complex image classification tasks. This work therefore highlights the potential of VCSEL-based platforms for novel, ultrafast, all-optical neuromorphic processors interfacing with current computation and communication systems for use in future light-enabled AI and computer vision functionalities.

A New Spiking Convolutional Recurrent Neural Network (SCRNN) With Applications to Event-Based Hand Gesture Recognition.

The combination of neuromorphic visual sensors and spiking neural network offers a high efficient bio-inspired solution to real-world applications. However, processing event- based sequences remains challenging because of the nature of their asynchronism and sparsity behavior. In this paper, a novel spiking convolutional recurrent neural network (SCRNN) architecture that takes advantage of both convolution operation and recurrent connectivity to maintain the spatial and temporal relations from event-based sequence data are presented. The use of recurrent architecture enables the network to have a sampling window with an arbitrary length, allowing the network to exploit temporal correlations between event collections. Rather than standard ANN to SNN conversion techniques, the network utilizes a supervised Spike Layer Error Reassignment (SLAYER) training mechanism that allows the network to adapt to neuromorphic (event-based) data directly. The network structure is validated on the DVS gesture dataset and achieves a 10 class gesture recognition accuracy of 96.59% and an 11 class gesture recognition accuracy of 90.28%.

Perception Understanding Action: Adding Understanding to the Perception Action Cycle With Spiking Segmentation.

Traditionally the Perception Action cycle is the first stage of building an autonomous robotic system and a practical way to implement a low latency reactive system within a low Size, Weight and Power (SWaP) package. However, within complex scenarios, this method can lack contextual understanding about the scene, such as object recognition-based tracking or system attention. Object detection, identification and tracking along with semantic segmentation and attention are all modern computer vision tasks in which Convolutional Neural Networks (CNN) have shown significant success, although such networks often have a large computational overhead and power requirements, which are not ideal in smaller robotics tasks. Furthermore, cloud computing and massively parallel processing like in Graphic Processing Units (GPUs) are outside the specification of many tasks due to their respective latency and SWaP constraints. In response to this, Spiking Convolutional Neural Networks (SCNNs) look to provide the feature extraction benefits of CNNs, while maintaining low latency and power overhead thanks to their asynchronous spiking event-based processing. A novel Neuromorphic Perception Understanding Action (PUA) system is presented, that aims to combine the feature extraction benefits of CNNs with low latency processing of SCNNs. The PUA utilizes a Neuromorphic Vision Sensor for Perception that facilitates asynchronous processing within a Spiking fully Convolutional Neural Network (SpikeCNN) to provide semantic segmentation and Understanding of the scene. The output is fed to a spiking control system providing Actions. With this approach, the aim is to bring features of deep learning into the lower levels of autonomous robotics, while maintaining a biologically plausible STDP rule throughout the learned encoding part of the network. The network will be shown to provide a more robust and predictable management of spiking activity with an improved thresholding response. The reported experiments show that this system can deliver robust results of over 96 and 81% for accuracy and Intersection over Union, ensuring such a system can be successfully used within object recognition, classification and tracking problem. This demonstrates that the attention of the system can be tracked accurately, while the asynchronous processing means the controller can give precise track updates with minimal latency.

Convolutional Neural Networks for Pathological Voice Detection.

Acoustic analysis using signal processing tools can be used to extract voice features to distinguish whether a voice is pathological or healthy. The proposed work uses spectrogram of voice recordings from a voice database as the input to a Convolutional Neural Network (CNN) for automatic feature extraction and classification of disordered and normal voice. The novel classifier achieved 88.5%, 66.2% and 77.0% accuracy on training, validation and testing data set respectively on 482 normal and 482 organic dysphonia speech files. It reveals that the proposed novel algorithm on the Saarbruecken Voice Database can effectively been used for screening pathological voice recordings.

Study on Interaction Between Temporal and Spatial Information in Classification of EMG Signals for Myoelectric Prostheses.

Advanced forearm prosthetic devices employ classifiers to recognize different electromyography (EMG) signal patterns, in order to identify the user's intended motion gesture. The classification accuracy is one of the main determinants of real-time controllability of a prosthetic limb and hence the necessity to achieve as high an accuracy as possible. In this paper, we study the effects of the temporal and spatial information provided to the classifier on its off-line performance and analyze their inter-dependencies. EMG data associated with seven practical hand gestures were recorded from partial-hand and trans-radial amputee volunteers as well as able-bodied volunteers. An extensive investigation was conducted to study the effect of analysis window length, window overlap, and the number of electrode channels on the classification accuracy as well as their interactions. Our main discoveries are that the effect of analysis window length on classification accuracy is practically independent of the number of electrodes for all participant groups; window overlap has no direct influence on classifier performance, irrespective of the window length, number of channels, or limb condition; the type of limb deficiency and the existing channel count influence the reduction in classification error achieved by adding more number of channels; partial-hand amputees outperform trans-radial amputees, with classification accuracies of only 11.3% below values achieved by able-bodied volunteers.

3-dimensional throat region segmentation from MRI data based on Fourier interpolation and 3-dimensional level set methods.

A new algorithm for 3D throat region segmentation from magnetic resonance imaging (MRI) is presented. The proposed algorithm initially pre-processes the MRI data to increase the contrast between the throat region and its surrounding tissues and to reduce artifacts. Isotropic 3D volume is reconstructed using the Fourier interpolation. Furthermore, a cube encompassing the throat region is evolved using level set method to form a smooth 3D boundary of the throat region. The results of the proposed algorithm on real and synthetic MRI data are used to validate the robustness and accuracy of the algorithm.

Automatic misclassification rejection for LDA classifier using ROC curves.

This paper presents a technique to improve the performance of an LDA classifier by determining if the predicted classification output is a misclassification and thereby rejecting it. This is achieved by automatically computing a class specific threshold with the help of ROC curves. If the posterior probability of a prediction is below the threshold, the classification result is discarded. This method of minimizing false positives is beneficial in the control of electromyography (EMG) based upper-limb prosthetic devices. It is hypothesized that a unique EMG pattern is associated with a specific hand gesture. In reality, however, EMG signals are difficult to distinguish, particularly in the case of multiple finger motions, and hence classifiers are trained to recognize a set of individual gestures. However, it is imperative that misclassifications be avoided because they result in unwanted prosthetic arm motions which are detrimental to device controllability. This warrants the need for the proposed technique wherein a misclassified gesture prediction is rejected resulting in no motion of the prosthetic arm. The technique was tested using surface EMG data recorded from thirteen amputees performing seven hand gestures. Results show the number of misclassifications was effectively reduced, particularly in cases with low original classification accuracy.